|
(11) | EP 4 442 196 A1 |
| (12) | EUROPEAN PATENT APPLICATION |
|
|
|
|
|||||||||||||||||||||||||||
| (54) | CIRRHOTIC PORTAL HYPERTENSION DIAGNOSING METHOD, APPARATUS, DEVICE, AND MEDIUM |
| (57) The present disclosure relates to the technical field of medical diagnosis, and particularly
provides cirrhotic portal hypertension diagnosing method, apparatus, device, and medium,
which process an input MRI image or CT image using a trained liver vessel three-dimensional
segmentation model, to obtain a liver contour image of a patient, wherein the liver
contour image includes a portal vascular tree, a hepatic vein vascular tree, an aortic
vascular tree, and an inferior vena cava vascular tree; automatically extract corresponding
vascular geometric parameters from each vascular tree in the obtained liver contour
image; process input vascular geometric parameters using the trained cirrhotic portal
hypertension diagnostic model, to obtain a cirrhotic portal hypertension diagnosis
result of the patient. Thus, the cirrhotic portal hypertension is diagnosed through
the vascular geometric characteristics, so that on one hand, the diagnosis result
is more accurate, and on the other hand, powerful pathophysiological explanations
can be supplemented to a model result. |
Technical Field
[0001] The present disclosure relates to the technical field of medical diagnosis, and in particular, to a cirrhotic portal hypertension diagnosing method, an apparatus, a device, and a medium.
Background Art
[0002] Portal hypertension non-invasive detection methods include serological examination, anatomical image marker, and physical substitution method based on histological characteristics. Currently, there still lacks a non-invasive gold standard that can replace HVPG for portal hypertension risk stratification and monitoring therapeutic effects. The main problem thereof lies in that invasive detection of portal hypertension is restricted in clinical applications, while the non-invasive risk stratification method of portal hypertension still needs to be further studied and improved: the evaluation efficacy is insufficient, and the invasive gold standard cannot be replaced; and there is insufficient generalization performance for different causes and races.
[0003] Although the technique proposed in the prior art for evaluating portal hypertension based on radiomics has achieved certain breakthroughs in evaluation efficacy, this technique is limited in research errors due to different causes, populations, and heterogeneity of examination machines, and inaccurate results caused by model establishment based on a two-dimensional plane and insufficiently optimized calculation parameters.
Summary
[0004] In view of this, the present disclosure aims at providing cirrhotic portal hypertension diagnosing method, apparatus, device, and medium, which can realize precise non-invasive diagnosis of portal hypertension.
[0005] In a first aspect, the present disclosure provides a cirrhotic portal hypertension diagnosing method, wherein the method includes steps of:
processing an input MRI image or CT image using a trained liver vessel three-dimensional segmentation model, so as to obtain a liver contour image of a patient, wherein the liver contour image includes a portal vascular tree, a hepatic vein vascular tree, an aortic vascular tree, and an inferior vena cava vascular tree;
extracting automatically corresponding vascular geometric parameters from each vascular tree in the obtained liver contour image; and
processing input vascular geometric parameters using a trained cirrhotic portal hypertension diagnostic model, so as to obtain a cirrhotic portal hypertension diagnosis result of the patient.
[0006] In some embodiments, the liver vessel three-dimensional segmentation model adopts a U-Net network segmentation architecture, and before extracting features from the input MRI image or CT image, the liver vessel three-dimensional segmentation model further performs pre-processing, including steps of:
performing field-of-view cropping on the input MRI image or CT image according to a pre-set dimension;
performing resolution re-sampling on the MRI image or CT image having undergone the field-of-view cropping; and
slicing the re-sampled MRI image or CT image by means of trilinear interpolation, and normalizing signal strength using z-score, so as to obtain a pre-processed enhanced image.
[0007] In some embodiments, a coding path of the U-Net in the liver vessel three-dimensional segmentation model contains 5~20 convolution layers and pooling layers, each layer contains a 3×3 convolution kernel and a rectified linear unit activation function, and a 3×3 convolution layer with a stride of 2 is connected immediately after a first convolution layer.
[0008] In some embodiments, for the CT image, before a pre-processing pipeline is applied, a signal intensity window is set to be [-200,200]HU.
[0009] In some embodiments, the vascular geometric parameters include one or more of vessel volume; vessel volume percentage; the number of vessel branch nodes; the number of vessel terminal nodes; the number of vessel branches; whole vessel length; vessel main branch length; vessel sub-branch length; vessel main branch curvature; vessel sub-branch curvature; vessel main branch tortuosity; vessel sub-branch tortuosity; equivalent diameter, minimum diameter, and roundness of vessel main branch; and first sub-branch angle.
[0010] In some embodiments, the step of extracting automatically corresponding vascular geometric parameters from each vascular tree in the obtained liver contour image includes steps of:
identifying a portal vascular tree, a hepatic vein vascular tree, an aortic vascular tree, and an inferior vena cava vascular tree from the obtained liver contour image;
sampling central lines of the portal vascular tree, the hepatic vein vascular tree, the aortic vascular tree, and the inferior vena cava vascular tree, respectively; and
calculating corresponding vascular geometric parameters based on the central lines of the portal vascular tree, the hepatic vein vascular tree, the aortic vascular tree, and the inferior vena cava vascular tree.
[0011] In some embodiments, vessel tortuosity is calculated as follows: counting each branch of the central line of vessel, calculating an Euclidean distance and a curve distance thereof, and dividing the curve distance by the Euclidean distance and subtracting 1, so that the tortuosity of the branch is obtained.
[0012] In some embodiments, the cirrhotic portal hypertension diagnostic model uses a support vector machine for binary classification, and an output cirrhotic portal hypertension diagnosis result is HVPG normal or HVPG abnormal.
[0013] In some embodiments, a pre-constructed cirrhotic portal hypertension diagnostic model is trained in a manner as follows, including steps of:
obtaining several sets of vascular geometric parameters as a data set using a trained liver vessel three-dimensional segmentation model;
labelling the data set with a true cirrhotic portal hypertension diagnosis result, and dividing the data set into a training set, a verification set, and a test set according to a set ratio;
training the pre-constructed cirrhotic portal hypertension diagnostic model based on the training set, and performing parameter adjustment on the pre-constructed cirrhotic portal hypertension diagnostic model based on the verification set, until parameters of the cirrhotic portal hypertension diagnostic model converge, so as to obtain a to-be-tested cirrhotic portal hypertension diagnostic model; and
evaluating the to-be-tested cirrhotic portal hypertension diagnostic model based on the test set, wherein if an evaluation result reaches a set threshold, the to-be-tested cirrhotic portal hypertension diagnostic model is determined as the trained cirrhotic portal hypertension diagnostic model.
[0014] In some embodiments, a pre-set ratio of the training set, the verification set, and the test set is 6:2:2.
[0015] In some embodiments, an Adam optimizer is selected to train the pre-constructed liver vessel three-dimensional segmentation model.
[0016] In some embodiments, an initial learning rate is set to be 0.0001, a batch size is set to be 4, an echo number is 100, and U-Net learns a consistent pattern from a training set composed of reference images with annotations, and then predicts images in all remaining cases.
[0017] An embodiment of the present disclosure provides a cirrhotic portal hypertension diagnosing apparatus, wherein the apparatus includes:
a processing module, configured to process an input MRI image or CT image using a trained liver vessel three-dimensional segmentation model, so as to obtain a liver contour image of a patient, wherein the liver contour image includes a portal vascular tree, a hepatic vein vascular tree, an aortic vascular tree, and an inferior vena cava vascular tree;
a calculating module, configured to automatically extract corresponding vascular geometric parameters from each vascular tree in the obtained liver contour image; and
a diagnosing module, configured to process input vascular geometric parameters using a trained cirrhotic portal hypertension diagnostic model, so as to obtain a cirrhotic portal hypertension diagnosis result of the patient.
[0018] An embodiment of the present disclosure provides an electronic device, including a processor, a memory, and a bus, wherein the memory stores a machine readable instruction executable by the processor, and when the electronic device is running, the processor is in communication with the memory via the bus, and the machine readable instruction, when executed by the processor, executes the steps of the cirrhotic portal hypertension diagnosing method according to any one of the above.
[0019] An embodiment of the present disclosure provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and the computer program, when run by a processor, executes the steps of the cirrhotic portal hypertension diagnosing method according to any one of the above.
[0020] The cirrhotic portal hypertension diagnosing method, apparatus, device, and medium in the present disclosure process the input MRI image or CT image using the trained liver vessel three-dimensional segmentation model, so as to obtain the liver contour image of the patient, wherein the liver contour image includes the portal vascular tree, the hepatic vein vascular tree, the aortic vascular tree, and the inferior vena cava vascular tree; automatically extract corresponding vascular geometric parameters from each vascular tree in the obtained liver contour image; process the input vascular geometric parameters using the trained cirrhotic portal hypertension diagnostic model, so as to obtain the cirrhotic portal hypertension diagnosis result of the patient. Thus, the cirrhotic portal hypertension is diagnosed through the vascular geometric characteristics, so that on one hand, the diagnosis result is more accurate, and on the other hand, powerful pathophysiological explanations can be supplemented to a model result.
Brief Description of Drawings
[0021] In order to illustrate the technical solutions of embodiments of the present disclosure more clearly, drawings that need to be used in the embodiments are introduced briefly below, and it should be understood that the following drawings merely shown some embodiments of the present disclosure, and therefore should not be construed as limitation to the scope, and a person ordinarily skilled in the art still could obtain other relevant drawings according to these drawings without using any creative efforts.
FIG. 1 shows a flowchart of a cirrhotic portal hypertension diagnosing method provided in an embodiment of the present disclosure;
FIG. 2 shows a schematic diagram of processing an input MRI image or CT image using a trained liver vessel three-dimensional segmentation model in an embodiment of the present disclosure;
FIG. 3 shows a schematic diagram of extracting automatically corresponding vascular geometric parameters from each vascular tree in the obtained liver contour image in an embodiment of the present disclosure;
FIG. 4 shows a schematic diagram of obtaining a cirrhotic portal hypertension diagnosis result of a patient using a trained cirrhotic portal hypertension diagnostic model in an embodiment of the present disclosure;
FIG 5 shows a structural block diagram of a cirrhotic portal hypertension diagnosing apparatus provided in an embodiment of the present disclosure; and
FIG. 6 shows a structural block diagram of an electronic device provided in an embodiment of the present disclosure.
Detailed Description of Embodiments
[0022] In order to make objectives, technical solutions, and advantages of the embodiments of the present disclosure clearer, the technical solutions in the embodiments of the present disclosure will be described clearly and completely below in conjunction with drawings in the embodiments of the present disclosure. It should be understood that the drawings in the present disclosure are merely for the illustrative and descriptive purpose, rather than limiting the scope of protection of the present disclosure. Besides, it should be understood that the schematic drawings are not drawn to scale. The flowcharts used in the present disclosure show operations implemented according to some embodiments of the present disclosure. It should be understood that the operations of the flowcharts may be implemented out of order, and steps without contextual logic may be reversed in order or simultaneously implemented. In addition, one skilled in the art, guided by the present disclosure, could add one or more other operations to the flowcharts, or remove one or more operations from the flowcharts.
[0023] Besides, some but not all embodiments of the present disclosure are described. Generally, components in the embodiments of the present disclosure, as described and shown in the drawings herein, may be arranged and designed in various different configurations. Therefore, the detailed description below of the embodiments of the present disclosure provided in the drawings is not intended to limit the claimed scope of the present disclosure, but merely illustrates chosen embodiments of the present disclosure.
[0024] It should be noted that the term "include" will be used in the embodiments of the present disclosure to indicate the existence of features specified thereafter, but not to exclude addition of other features.
[0025] In view of the technical problem proposed in the background art, the present disclosure provides cirrhotic portal hypertension diagnosing method, apparatus, device, and medium, which can realize precise and non-invasive diagnosis of portal hypertension.
[0026] In order to facilitate understanding the present embodiment, the cirrhotic portal hypertension diagnosing method, apparatus, device, and medium provided in the embodiments of the present disclosure are introduced in detail below. The above method provided in the embodiments of the present disclosure can be applied to a terminal device, and also can be applied to a server, wherein when the above method is applied to a server, it may be cloud medical treatment, i.e., a man-machine interaction interface is provided via a terminal device, the terminal device sends an operation instruction triggered by a user via a man-machine interaction screen to the server, and the server performs data processing in response to the user's operation instruction and returns a post-processing result (vascular geometric parameters and cirrhotic portal hypertension diagnosis) to the terminal device.
[0027] Referring to FIG. 1 of the description, an embodiment of the present disclosure provides a cirrhotic portal hypertension diagnosing method, wherein the method includes the following steps:
S1, processing an input MRI image or CT image using a trained liver vessel three-dimensional segmentation model, so as to obtain a liver contour image of a patient, wherein the liver contour image includes a portal vascular tree, a hepatic vein vascular tree, an aortic vascular tree, and an inferior vena cava vascular tree.
[0028] With reference to FIG. 2, in step S1, the liver vessel three-dimensional segmentation model adopts a U-Net network segmentation architecture, and before extracting features from the input MRI image or CT image, the liver vessel three-dimensional segmentation model further performs pre-processing, specifically including steps of: performing field-of-view cropping on the input MRI image or CT image according to a pre-set dimension; performing resolution re-sampling on the MRI image or CT image having undergone the field-of-view cropping; and slicing the re-sampled MRI image or CT image by means of trilinear interpolation, and normalizing signal strength using z-score, so as to obtain a pre-processed enhanced image.
[0029] In some embodiments, a coding path of the U-Net in the liver vessel three-dimensional segmentation model contains 5~20 convolution layers and pooling layers, each layer contains a 3×3 convolution kernel and a rectified linear unit (ReLu) activation function, and a 3×3 convolution layer with a stride of 2 is connected immediately after a first convolution layer. In the above, a joint feature is provided for two tasks by using a connection between two paths, so as to improve prediction performance. When pre-processing the patient's MRI image or CT image, the field of view is first cropped into 640×640 mm2; image resampling resolution is 0.625×0.625 mm2, and a size of a corresponding matrix is 512×512; a slice thickness is interpolated to 0.625 mm by means of trilinear interpolation; the signal intensity is normalized using the z-score, so as to obtain the enhanced image, particularly for a CT image, before a pre-processing pipeline is applied, a signal intensity window is set to be [-200,200]HU. Furthermore, the acquired enhanced image is input into the trained liver vessel three-dimensional segmentation model to extract the feature, thus obtaining the patient's liver contour image. In this embodiment, the obtained liver contour image includes the portal vascular tree, the hepatic vein vascular tree, the aortic vascular tree, and the inferior vena cava vascular tree.
[0030] It should be noted that, when training the pre-constructed liver vessel three-dimensional segmentation model, by selecting a DICE similarity coefficient as a loss function, a network can be effectively trained. In an embodiment, an Adam optimizer is selected to perform network training, wherein an initial learning rate is set to be 0.0001, a batch size is set to be 4, an echo number is 100, and U-Net learns a consistent pattern from a training set composed of reference images with annotations, and then predicts images in all remaining cases. Training time of the model is about 15 h, and testing time of each case is within 10 s. In the above, the process of training the pre-constructed liver vessel three-dimensional segmentation model should be a technical means well known to a person skilled in the art, and will not be repeated herein.
[0031] S2, extracting automatically corresponding vascular geometric parameters from each vascular tree in the obtained liver contour image.
[0032] With reference to FIG. 3 of the description, in step S2, when calculating the vascular geometric parameters, first, the portal vascular tree, the hepatic vein vascular tree, the aortic vascular tree, and the inferior vena cava vascular tree, i.e. branch levels of vessel, should be identified from the liver contour image acquired in step S1. For example, a central line is extracted from a segmented vessel based on a decision-tree method, a graph representation of vascular system is constructed according to the extracted central line, and further automatic identification is performed according to features of the central lines of different vascular trees. In the present disclosure, the vascular geometric parameters include sixteen parameters in total, namely, vessel volume; vessel volume percentage; number of vessel branch nodes; number of vessel terminal nodes; number of vessel branches; whole vessel length; vessel main branch length; vessel sub-branch length; vessel main branch curvature; vessel sub-branch curvature; vessel main branch tortuosity; vessel sub-branch tortuosity; equivalent diameter, minimum diameter, and roundness of vessel main branch; and first sub-branch angle. With these sixteen parameters, powerful pathophysiological explanations can be provided for subsequent cirrhotic portal hypertension diagnosis, and the diagnostic accuracy can be improved.
[0033] In this embodiment, when calculating the vascular geometric parameters, the central line can be appropriately sampled, and then parameters are calculated, so as to improve accuracy of the parameters and avoid influence of extreme values. Specifically:
vessel volume (ml)=voxel_count·unit_vol (volume=voxel_count×unit_vol), where voxel_count is the total number of pixel values greater than zero in the image, and unit_vol=spacing[0]*spacing[1]*spacing[2]*0.001 (unit_vol=spacing[0]*spacing[1]*spacing[2]*0.001);
vessel volume percentage=intrahepatic vessel volume/liver volume;
the number of vessel terminal nodes (terminal_node_num): traversing the central lines of the vessel, recording coordinates of a starting point and an ending point of each central line, and counting the number after getting rid of the same coordinates, so as to obtain the number of terminal nodes;
the number of vessel branch nodes (branch_node_num): traversing the central lines of the vessel, recording coordinates of bifurcations of the central lines, and counting the number of bifurcations;
the number of vessel branches (branch_num): branch_num =term inal_node _num +branch_node _num -1 ;
vessel sub-branch length (branch_length): calculating length of the central line of each sub-branch;
vessel main branch length (main_length): identifying the central line corresponding to a main branch, and calculating the length;
whole vessel length (whole_length): summing lengths of all vessel branches;
vessel curvature calculation: the central line of vessel and a point p thereon, a circle being tangent to the central line at the point p, the circle being an osculating circle, and reciprocal of radius of the osculating circle being curvature of the point p;
vessel main branch curvature: counting each point in the central line of the vessel main branch, and calculating the curvatures and taking a mean value;
vessel sub-branch curvature: counting each point in the central line of the vessel sub-branch, and calculating the curvatures and taking a mean value;
calculation of vessel tortuosity: counting each branch of the central line of vessel,
calculating an Euclidean distance (Euclid_dist) and a curve distance (curve_dist)
thereof, and dividing the curve distance by the Euclidean distance and subtracting
1, so as to obtain the tortuosity of the branch (Tortuosity);
vessel main branch tortuosity: calculating an Euclidean distance and a curve distance of the central line of the vessel main branch, and calculating the tortuosity;
vessel sub-branch tortuosity: calculating an Euclidean distance and a curve distance of the central line of each vessel sub-branch, and calculating the tortuosity and taking a mean value;
equivalent diameter, minimum diameter, and roundness of vessel main branch: a cross section of a point corresponding to the vessel needs to be used for calculating this parameter, where a normal vector p0p1 of a plane is formed by two adjacent points Po, P in the central line, and then a cross section area "area" and a cross section perimeter "perimeter" of the vessel of this plane at this point are calculated;
then, the equivalent diameter is:
;
the minimum diameter is min_radius=diameter of maximum inscribed circle of a cross section at this point;
the roundness is:
; then the roundness of this cross section is calculated, where area is area of the
cross section, and perimeter is perimeter of the cross section;
the first sub-branch angle: finding a sub-branch node P0 at a vessel main branch, and midpoints P1, P2, Ps...of adjacent sub-branches, then respectively calculating an included angle θ
of vectors P0P1 and P0P2, and so on, and finally, calculating a mean value of θ0, θ1...
[0034] S3, processing the input vascular geometric parameters using a trained cirrhotic portal hypertension diagnostic model, so as to obtain a cirrhotic portal hypertension diagnosis result of the patient.
[0035] With reference to FIG. 4 of the description, in step S3, the cirrhotic portal hypertension diagnostic model uses a support vector machine for binary classification, and uses a non-linear kernel function, so as to output a cirrhotic portal hypertension diagnosis result of HVPG normal or HVPG abnormal according to the input vascular geometric parameters.
[0036] In the above, when training the pre-constructed cirrhotic portal hypertension diagnostic model, the trained liver vessel three-dimensional segmentation model is used to obtain several sets of vascular geometric parameters as a data set; the data set is labeled with a true cirrhotic portal hypertension diagnosis result, and the data set is divided into a training set, a verification set, and a test set according to a set ratio; in an embodiment, a pre-set ratio of the training set, the verification set, and the test set is 6:2:2, then general parameters, such as weight and bias, of the pre-constructed cirrhotic portal hypertension diagnostic model are trained based on the training set, and performing parameter adjustment on the pre-constructed cirrhotic portal hypertension diagnostic model based on the verification set, for example, learning rate and the number of network layers are adjusted, until parameters of the cirrhotic portal hypertension diagnostic model converge, so as to obtain a to-be-tested cirrhotic portal hypertension diagnostic model; finally, the to-be-tested cirrhotic portal hypertension diagnostic model is evaluated based on the test set, wherein if an evaluation result reaches a set threshold, for example, the set threshold is an accuracy rate of 0.9, the to-be-tested cirrhotic portal hypertension diagnostic model is determined as the trained cirrhotic portal hypertension diagnostic model. In the above, the process of training the pre-constructed cirrhotic portal hypertension diagnostic model should be a technical means well known to a person skilled in the art, and will not be repeated here.
[0037] Instead of directly inputting the liver contour image output by the liver vessel three-dimensional segmentation model into the cirrhotic portal hypertension diagnostic model to acquire the cirrhotic portal hypertension diagnosis result of the patient, the cirrhotic portal hypertension diagnosing method provided in the present disclosure firstly processes the liver contour image, and extract the vascular geometric parameters from each vascular tree in the liver contour image, then inputs the vascular geometric parameters into the cirrhotic portal hypertension diagnostic model, so as to acquire the cirrhotic portal hypertension diagnosis result of the patient, so that on one hand, the diagnosis result is more accurate, and on the other hand, powerful pathophysiological explanations can be supplemented to the diagnosis result. In addition, the liver vessel three-dimensional segmentation model used is based on an MRI/CT multi-modal image, which is beneficial to clinical application and popularization.
[0038] Based on the same inventive concept, an embodiment of the present disclosure further provides a cirrhotic portal hypertension diagnosing apparatus, and as the principle for solving the problem by the apparatus in the embodiment of the present disclosure is similar to that of the cirrhotic portal hypertension diagnosing method in the embodiments of the present disclosure, reference can be made to implementation of the method for implementation of the apparatus, and repetition will not be made herein.
[0039] As shown in FIG. 5 of the description, the present disclosure further provides a cirrhotic portal hypertension diagnosing apparatus, wherein the apparatus includes:
a processing module 501, configured to process an input MRI image or CT image using a trained liver vessel three-dimensional segmentation model, so as to obtain a liver contour image of a patient, wherein the liver contour image includes a portal vascular tree, a hepatic vein vascular tree, an aortic vascular tree, and an inferior vena cava vascular tree;
a calculating module 502, configured to automatically extract corresponding vascular geometric parameters from each vascular tree in the obtained liver contour image; and
a diagnosing module 503, configured to process input vascular geometric parameters using a trained cirrhotic portal hypertension diagnostic model, so as to obtain a cirrhotic portal hypertension diagnosis result of the patient.
[0040] In some embodiments, the liver vessel three-dimensional segmentation model adopts a u-net network segmentation architecture, and the processing module 501 further performs pre-processing on features of the MRI image or CT image input into the trained liver vessel three-dimensional segmentation model, including:
performing field-of-view cropping on the input MRI image or CT image according to a pre-set dimension;
performing resolution re-sampling on the MRI image or CT image having undergone the field-of-view cropping; and
slicing the re-sampled MRI image or CT image by means of trilinear interpolation, and normalizing signal strength using z-score, so as to obtain a pre-processed enhanced image.
[0041] In some embodiments, extracting automatically corresponding vascular geometric parameters from each vascular tree in the obtained liver contour image by the calculating module 502 includes:
identifying a portal vascular tree, a hepatic vein vascular tree, an aortic vascular tree, and an inferior vena cava vascular tree from the obtained liver contour image;
sampling central lines of the portal vascular tree, the hepatic vein vascular tree, the aortic vascular tree, and the inferior vena cava vascular tree, respectively; and
calculating corresponding vascular geometric parameters based on the central lines of the portal vascular tree, the hepatic vein vascular tree, the aortic vascular tree, and the inferior vena cava vascular tree.
[0042] In some embodiments, the diagnosing module 503 further trains a pre-constructed cirrhotic portal hypertension diagnostic model, including:
obtaining several sets of vascular geometric parameters as a data set using a trained liver vessel three-dimensional segmentation model;
labelling the data set with a true cirrhotic portal hypertension diagnosis result, and dividing the data set into a training set, a verification set, and a test set according to a set ratio;
training the pre-constructed cirrhotic portal hypertension diagnostic model based on the training set, and performing parameter adjustment on the pre-constructed cirrhotic portal hypertension diagnostic model based on the verification set, until parameters of the cirrhotic portal hypertension diagnostic model converge, so as to obtain a to-be-tested cirrhotic portal hypertension diagnostic model; and
evaluating the to-be-tested cirrhotic portal hypertension diagnostic model based on the test set, wherein if an evaluation result reaches a set threshold, the to-be-tested cirrhotic portal hypertension diagnostic model is determined as the trained cirrhotic portal hypertension diagnostic model.
[0043] The cirrhotic portal hypertension diagnosing apparatus provided in the present disclosure processes the input MRI image or CT image using the trained liver vessel three-dimensional segmentation model through the processing module, so as to obtain the liver contour image of the patient, wherein the liver contour image includes the portal vascular tree, the hepatic vein vascular tree, the aortic vascular tree, and the inferior vena cava vascular tree; automatically extracts corresponding vascular geometric parameters from each vascular tree in the obtained liver contour image through the calculating module; process the input vascular geometric parameters using the trained cirrhotic portal hypertension diagnostic model through the diagnosing module, so as to obtain the cirrhotic portal hypertension diagnosis result of the patient. Thus, the cirrhotic portal hypertension is diagnosed through the vascular geometric characteristics, so that on one hand, the diagnosis result is more accurate, and on the other hand, powerful pathophysiological explanations can be supplemented to a model result.
[0044] Based on the same concept of the present disclosure, as shown in FIG. 6 of the description, a structure of an electronic device 600 is provided in an embodiment of the present disclosure, the electronic device 600 includes: at least one processor 601, at least one network interface 604 or other user interfaces 603, a memory 605, and at least one communication bus 602. The communication bus 602 is configured to enable connection communication between these components. This electronic device 600 optionally includes the user interface 603, including a display (for example, a touchscreen, an LCD, a CRT, a holographic imager or a projector), a keyboard, or a click device (for example, a mouse, a trackball, a touch panel, or a touchscreen).
[0045] The memory 605 may include a read-only memory and a random access memory, and provides an instruction and data to the processor 601. A part of the memory 605 may also include a non-volatile random access memory (NVRAM).
[0046] In some embodiments, the memory 605 stores the following elements: a protectable module or a data structure, or a subset thereof, or an extension set thereof:
an operating system 6051, containing various system programs, configured to realize various basic services and process hardware-based tasks; and
an application module 6052, containing various applications, such as a launcher, a media player, and a browser, configured to realize various application services.
[0047] In the embodiments of the present disclosure, by invoking a program or an instruction stored in the memory 605, the processor 601 is configured for executing steps in the cirrhotic portal hypertension diagnosing method according to the present disclosure, and can realize precise and non-invasive diagnosis of portal hypertension.
[0048] The present disclosure further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and the computer program, when run by a processor, executes the steps in the cirrhotic portal hypertension diagnosing method.
[0049] Specifically, the storage medium can be a general-purpose storage medium, for example, removable disk and hard disk, and when the computer program on the storage medium is run, the above cirrhotic portal hypertension diagnosing method can be executed.
[0050] In the embodiments provided in the present disclosure, it should be understood that the apparatus and the method disclosed may be implemented in other manners. The apparatus embodiment described in the above is merely exemplary, for example, the units are merely divided according to logical functions, but they may be divided in other manners in practical implementation, for another example, multiple units or components may be combined or may be integrated into another system, or some features may be omitted, or not executed. In addition, mutual couplings or direct coupling or communication connection as shown or discussed may be indirect coupling or communication connection via some communication interfaces, means or units, and may be in an electrical form, a mechanical form or other forms.
[0051] The units described as separate parts may be or also may not be physically separated, the parts displayed as units may be or also may not be physical units, i.e., they may be located at one place, or also may be distributed on a plurality of network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution in the present embodiment.
[0052] Besides, various functional units in the embodiments of the present disclosure may be integrated into one processing unit, or each unit also may exist in a physically independent way, or two or more than two units may be integrated into one unit.
[0053] If a function is realized in a form of software functional unit and is sold or used as an independent product, it may be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the present disclosure in essence or parts making contribution to the prior art or parts of the technical solutions can be embodied in form of a software product, and this computer software product is stored in a storage medium, including several instructions for making one computer device (which can be a personal computer, a server or a network device etc.) execute all or part of the steps of the methods of various embodiments of the present disclosure. The aforementioned storage medium includes various media in which program codes can be stored, such as U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), magnetic disk or optical disk.
processing an input MRI image or CT image using a trained liver vessel three-dimensional segmentation model, so as to obtain a liver contour image of a patient, wherein the liver contour image comprises a portal vascular tree, a hepatic vein vascular tree, an aortic vascular tree, and an inferior vena cava vascular tree;
extracting automatically corresponding vascular geometric parameters from each vascular tree in the obtained liver contour image; and
processing input vascular geometric parameters using a trained cirrhotic portal hypertension diagnostic model, so as to obtain a cirrhotic portal hypertension diagnosis result of the patient.
performing field-of-view cropping on the input MRI image or CT image according to a pre-set dimension;
performing resolution re-sampling on the MRI image or CT image having undergone the field-of-view cropping; and
slicing the re-sampled MRI image or CT image by means of trilinear interpolation, and normalizing signal strength using z-score, so as to obtain a pre-processed enhanced image.
identifying a portal vascular tree, a hepatic vein vascular tree, an aortic vascular tree, and an inferior vena cava vascular tree from the obtained liver contour image;
sampling central lines of the portal vascular tree, the hepatic vein vascular tree, the aortic vascular tree, and the inferior vena cava vascular tree, respectively; and
calculating corresponding vascular geometric parameters based on the central lines of the portal vascular tree, the hepatic vein vascular tree, the aortic vascular tree, and the inferior vena cava vascular tree.
obtaining several sets of vascular geometric parameters as a data set using a trained liver vessel three-dimensional segmentation model;
labelling the data set with a true cirrhotic portal hypertension diagnosis result, and dividing the data set into a training set, a verification set, and a test set according to a set ratio;
training the pre-constructed cirrhotic portal hypertension diagnostic model based on the training set, and performing parameter adjustment on the pre-constructed cirrhotic portal hypertension diagnostic model based on the verification set, until parameters of the cirrhotic portal hypertension diagnostic model converge, so as to obtain a to-be-tested cirrhotic portal hypertension diagnostic model; and
evaluating the to-be-tested cirrhotic portal hypertension diagnostic model based on the test set, wherein if an evaluation result reaches a set threshold, the to-be-tested cirrhotic portal hypertension diagnostic model is determined as the trained cirrhotic portal hypertension diagnostic model.
a processing module, configured to process an input MRI image or CT image using a trained liver vessel three-dimensional segmentation model,
so as to obtain a liver contour image of a patient, wherein the liver contour image comprises a portal vascular tree, a hepatic vein vascular tree, an aortic vascular tree, and an inferior vena cava vascular tree;
a calculating module, configured to automatically extract corresponding vascular geometric parameters from each vascular tree in the obtained liver contour image; and
a diagnosing module, configured to process input vascular geometric parameters using a trained cirrhotic portal hypertension diagnostic model,
so as to obtain a cirrhotic portal hypertension diagnosis result of the patient.